Glossary
Interview bias
Interview bias is any systematic error that makes an interviewer rate a candidate on something other than their ability to do the job.
Everyone who interviews has biases. The goal isn’t to find interviewers who don’t, but to design a process where those biases have less room to affect the outcome.
Common types
| Bias | What it looks like |
|---|---|
| First impression | The first two minutes set the rating, and the rest of the interview is spent confirming it. |
| Halo and horns | One strong trait (a famous employer) or one weak one (a nervous start) colors every other rating. |
| Affinity | Rating people higher because they went to the same school, share a hobby or talk like you. |
| Contrast | A decent candidate looks weak after an excellent one, or strong after a poor one. |
| Confirmation | Asking harder follow-ups of candidates you already doubt. |
| Central tendency | Rating everyone in the middle to avoid committing. |
| Fluency and accent | Mistaking polished delivery for competence, or an accent for weaker ability. |
Practical ways to reduce it
- Structure the interview. Same questions, same order. See structured interview.
- Use a rubric. Written anchors for each score level make it harder to rate on gut feel.
- Rate before discussing. Each interviewer scores independently, then the group talks.
- Require evidence. Every rating should point to something the candidate said or did.
- Separate what’s irrelevant. Photos, names of schools and hobbies rarely help a first-round decision.
- Review your outcomes. Compare ratings across groups where lawful. Patterns show where bias may be operating. See adverse impact.
An example
Two candidates give similar answers about handling an angry customer. One is a confident, fast talker. The other pauses, loses their train of thought once, then gives a detailed, specific account. Without a rubric, the first often gets the higher score. With one that asks for a specific situation, clear actions and a result, the second may well score higher, because they met more of the criteria.
AI and bias
AI can remove some human biases, like fatigue at the end of a long day or affinity with a candidate. It can also introduce new ones if it’s trained or instructed poorly. The questions to ask are practical: what inputs does it use, does it show its reasoning, and can people override it. LessRounds scores the transcript only, not faces or voices, and instructs the model never to mark anyone down for language, code-switching or grammar. No tool, human or AI, should be described as unbiased.